Fine-tuned small models predict x86-64 register inputs from outputs
Original title: BUSTA: Teaching Small Language Models to Reason About Binary Execution
GRPO fine-tuning lifts Nemotron-3-Nano knowable-register accuracy from 63.3% to 95.2%, within four points of frontier scores.
The BUSTA paper studies whether small open models can learn a reverse-engineering skill: given an x86-64 basic block and a desired output register state, predict the input register values that produce it. Fine-tuning Qwen3-1.7B and Nemotron-3-Nano with Group Relative Policy Optimization raises knowable-register accuracy from 73.2% to 85.3% and from 63.3% to 95.2%. Frontier models still score highest, up to 99.2%. The authors release both fine-tuned models and the training dataset.
- The task is inverse basic-block execution on x86-64.
- Qwen3-1.7B accuracy rises from 73.2% to 85.3%.
- Fine-tuned Nemotron-3-Nano reaches 95.2% versus frontier scores up to 99.2%.
- Both fine-tuned models and the training dataset are released.
From the original article
1 min read at arxiv.orgLarge Language Models (LLM) are increasingly applied to Reverse Engineering (RE) workflows. RE, however, differs from common LLM coding tasks as it requires models to reason about low-level program behavior. Furthermore, the deployment of tooling in resource-constrained sandboxed RE environments limits the ability to use frontier-scale solutions.
By Ritika Poddar · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.